CRII: CHS: RUI: Computational models of humans for studying and improving Human-AI interaction

CRII：CHS：RUI：用于研究和改善人机交互的人类计算模型

• 批准号: 2218226
• 负责人: Christopher Dancy
• 金额: $ 17.41万
• 依托单位: Pennsylvania State Univ University Park
• 依托单位国家: 美国
• 项目类别: Standard Grant
• 财政年份: 2022
• 资助国家: 美国
• 起止时间: 2022-01-15 至 2023-06-30
• 项目状态: 已结题
• 来源: https://www.nsf.gov/awardsearch/showAward?AWD_ID=2218226&HistoricalAwards=false
• 关键词: CRII; CHS; RUI; Computational; models

Understanding the interactions between humans and systems utilizing artificial intelligence (AI) requires an understanding of how human physiological changes impact memory and other mental processes. To realize the beneficial societal outcomes on, for example, interactions between humans and intelligent robots, it is important to develop simulations to test a variety of situations where memory under arousal states will function. In this project, the investigator articulates a research plan to use a simulation of the human mind and body to understand effects of physiological arousal on human memory and cognition and the consequences for human interaction with AI agents. Human-subject studies will be used to introduce stimuli to induce human arousal by using selected stressors and collect physiological and behavioral data during tasks that require cooperation between humans and AI agents. Societal benefits include an architecture to simulate a variety of human-AI interactions under various levels of arousal and stress. This architecture can be used to explore new ways to co-team humans with AI agents and set expectations for positive and/or negative behaviors that might occur in such collaborations. Undergraduate students at Bucknell University will be heavily involved as research assistants on this project. The investigator plans to develop simulations and elicit arousal states in human operators performing collaborative tasks with agents enabled by artificial intelligence (AI) to understand human-AI interaction. The goal is to better understand, through simulation, how algorithms for intelligent agents can be advanced and expanded to respond to human variations in behavior and memory processes under different levels of arousal, including levels that mimic stress. The objectives include developing simulations to examine contexts and tasks; understanding how environmental stimuli affect interactions between humans and intelligent agents; and determining how to advance algorithms to optimize human-AI cooperation and avoid maladaptive interactions. The investigator has already extended Adaptive Control of Thought-Rational (ACT-R) architecture to account for physiological influences on declarative and procedural memory. Physio-cognitive agents will be developed based on Bayesian and reinforcement learning to acquire knowledge of their environment. Human-AI task simulations will be implemented in a virtual environment to understand how arousal mediates intelligent behavior and how interaction with external environments that include AI agents may change performance, including through maladaptive behavior. The project seeks to discover computationally-enabled processes and contexts to amplify human capabilities. The resulting revised models and open-source code will be made available to the public for further explorations in human-AI interaction. The university is an undergraduate institution allowing significant participation of undergraduates in research.This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.

了解人类与使用人工智能（AI）的系统之间的相互作用需要了解人类生理变化如何影响记忆和其他心理过程。为了实现例如人类与智能机器人之间的相互作用的有益的社会成果，开发模拟以测试各种情况下的记忆将在唤醒状态下起作用很重要。在这个项目中，研究人员阐明了一项研究计划，以模拟人类的思想和身体，以了解生理唤醒对人类记忆和认知的影响，以及对与AI代理人相互作用的后果。人类受试者的研究将用于引入刺激，以使用选定的压力源来诱导人类唤醒，并在需要人与AI剂之间进行合作的任务期间收集生理和行为数据。社会利益包括一个建筑，以模拟各种唤醒和压力下的各种人类相互作用。该体系结构可用于探索与AI代理人共同团队人士的新方法，并设定对此类合作中可能发生的积极和/或负面行为的期望。巴克内尔大学的本科生将在该项目的研究助理中大量参与其中。研究人员计划开发模拟并引起人类操作员的唤醒状态，与人工智能（AI）启用的代理人执行协作任务，以了解人类的互动。目的是通过模拟更好地理解智能代理的算法如何进行推进和扩展，以应对在不同级别的唤醒水平下的行为和记忆过程中的变化，包括模仿压力的水平。 这些目标包括开发模拟以检查上下文和任务；了解环境刺激如何影响人与智能代理之间的相互作用；并确定如何推进算法以优化人类合作并避免适应不良的相互作用。研究者已经扩展了对思想理性（ACT-R）体系结构的自适应控制，以解释对声明性和程序记忆的生理影响。将根据贝叶斯和强化学习以获取对环境的了解，将开发生理认知剂。人类AI任务模拟将在虚拟环境中实施，以了解唤醒如何介导智能行为以及与包括AI代理在内的外部环境的相互作用如何改变性能，包括通过不良适应行为来改变性能。该项目旨在发现基于计算的过程和上下文，以扩大人类能力。由此产生的修订模型和开源代码将向公众提供，以进一步探索人类交互。该大学是一家本科机构，允许本科生在研究中大量参与。该奖项反映了NSF的法定任务，并被认为是使用基金会的知识分子优点和更广泛的影响审查标准的评估值得支持的。

项目成果

Examining the Effects of Race on Human-AI Cooperation

检查种族对人类与人工智能合作的影响

• DOI: 10.1007/978-3-030-80387-2_27
• 发表时间: 2021
• 期刊: & Prediction and Behavior Representation in Modeling and Simulation
• 作者: Atkins, A. A.;Brown, M. S.;Dancy, C. L.
• 通讯作者: Dancy, C. L.

The Forgotten margins of ai ethics

• DOI: 10.1145/3531146.3533157
• 发表时间: 2022-01-01
• 期刊: FAccT '22: 2022 ACM Conference on Fairness, Accountability, and Transparency
• 作者: Birhane, Abeba;Ruane, Elayne;Dancy, Christopher L.
• 通讯作者: Dancy, Christopher L.